Desenleştirilmiş Karma Verilerin Transfer Öğrenme Yöntemi Kullanılarak Evrişimli Sinir Ağlarıyla Sınıflandırılması

Abstract

Teknolojik ve bilimsel gelişmeler, farklı yapı ve boyuttaki veri setlerini görselleştirmeyi ve analiz etmeyi zorunlu hale getirmiştir. Veri görselleştirmede kullanılan grafikler hem tanımsal hem de analizleri destekleyici olarak keşifsel amaçlarla uygulanmaktadır. Grafikler, veri setlerindeki zaman, mekân, akış, ilişki, belirsizlik ve hiyerarşi gibi yapı ve olguları ortaya çıkarmak için de kullanılmaktadır. Günümüz araştırmalarında kullanılan veri setleri sadece nominal, ordinal ve / veya nümerik değişkenlerden (karma veri seti) oluşan yapısal veri setlerini içermemekte, yarı yapısal ya da yapısal olmayan veri setlerini de barındırmaktadır. Söz konusu veri setlerinde evrişimli sinir ağlarının başarısı literatürdeki birçok araştırmayla kanıtlanmıştır. Bu çalışma, yapısal veri setleri üzerinde evrişimli sinir ağlarını transfer öğrenme yöntemi ile eğiterek uygulayabilmek için bir yöntem önermektedir. Bu amaç doğrultusunda, literatürde çeşitli araştırmalarda kullanılan dokuz adet yapısal veri seti üzerinde önerilen yaklaşım uygulanarak, ağların ölçülen başarısı aynı veri setleriyle çalışılan literatürdeki diğer çalışmalarla karşılaştırılmış ve daha iyi sonuçlar elde edilmiştir.

Keywords

• Kategorik Değişken Kodlama
• Veri Görselleştirme
• Önceden Eğitilmiş Ağlar
• Transfer Öğrenme

Classification of Visualized Mixed Data With Convolutional Neural Networks Using Transfer Learning Method

Abstract

Technological and scientific advances have made it compulsory to visualize and analyze datasets of different structures and sizes. The graphics used in data visualization are applied for exploratory purposes, both in terms of definition and analysis. Graphs are also used to reveal structures and phenomena in datasets such as time, space, flow, relationships, uncertainty, and hierarchy. The datasets used in today's research not only contain structural datasets consisting of nominal, ordinary, and/or numerical variables (mixed datasets), but also contain semi-structural or non-structured data sets. Numerous studies in the literature have demonstrated the effectiveness of evolutionary neural networks in these datasets. This study suggests a method to train evolutionary neural networks to apply the transfer learning method to structural datasets. To this end, the proposed approach was applied to nine structural datasets used in various literature studies, comparing the measured success of the networks with other studies in the literature that studied the same data sets and achieving better results.

Keywords

• Categorical Variable Encoding
• Data Visualization
• Pre-Trained Networks
• Transfer Learning

References

1. [1] Dandıl, E., Polattimur, R. 2020. Dog behaviorrecognition and tracking based on faster R-CNN.Journal of the Faculty of Engineering andArchitecture of Gazi University, 35(2), 819-834.
2. [2] Yıldız, O. 2019. Melanoma detection fromdermoscopy images with deep learning methods:A comprehensive study. Journal of the Faculty ofEngineering and Architecture of Gazi University,34(4), 2241-2260.
3. [3]Pan, S. J., Yang, Q. 2010. A survey on transferlearning. IEEE Transactions on Knowledge andData Engineering, 22(10), 1345-1359.
4. [4]Zhou, Y., Nejati, H., Do, T. T., Cheung, N. M., Cheah,L.2016. Image-based vehicle analysis using deepneural network: A systematic study. IEEEInternational Conference on Digital SignalProcessing, 16-18 October, Beijing, 276-280.
5. [5]Krizhevsky, A., Sutskever, I., Hinton, G. E. 2012.ImageNet classification with deep convolutionalneural networks. NIPS'12: Proceedings of the25th International Conference on NeuralInformation Processing Systems, 3-6 December,Nevada, 1097–1105.
6. [6]Huang, G., Liu, Z., Maaten, L. V. D., Weinberger, K.Q.2018. Densely connected convolutionalnetworks. IEEE Conference on Computer Visionand Pattern Recognition, 21-26 July, Honolulu,4700-4708.
7. [7]He, K., Zhang, X., Ren, S., Sun, J. 2016. DeepResidual Learning for Image Recognition. IEEEConference on Computer Vision and PatternRecognition, 27-30 June, Las Vegas, 770-778.
8. [8]Theckedath, D., Sedamkar, R. R. 2020. Detectingaffect states using VGG16, ResNet50 andSE‑ResNet50 networks. Springer NatureComputer Science, 79, 1-7.

9. [9]Ye, H., Han, H., Zhu, L., Duan, Q. 2019. Vegetablepest image recognition method based onimproved VGG convolution neural network.Journal of Physics: Conference Series, 1237(3).
10. [10]Fırıldak, K., Talu, M. F. 2019. Evrişimsel sinirağlarında kullanılan transfer öğrenmeyaklaşımlarının incelenmesi. Anatolian Journal ofComputer Science, 4(2), 88-95.
11. [11]Lawrence, S., Giles, C. L., Tsoi, A. C., Back, A. D.1997. Face recognition: A convolutional neural-network approach. IEEE Transactions on NeuralNetworks, 8(1), 98-113.
12. [12]MathWorks. 2019. Learn About ConvolutionalNeural Networks.https://www.mathworks.com/help/deeplearning/ug/introduction-to-convolutional-neural-networks.html (Erişim Tarihi: 06.05.2023).
13. [13]Baykal, E., Doğan, H., Ercin, M. E., Ersoz, S., Ekinci,M.2019. Transfer learning with pre-trained deepconvolutional neural networks for serous cellclassification. Multimedia Tools and Applications,79, 15593-15611.
14. [14]Karabulut, E. M. 2016. Investigation of deeplearning approaches for biomedical dataclassification. Çukurova Üniversitesi, FenBilimleri Enstitüsü, Doktora Tezi, 108s, Adana.
15. [15]Goodfellow, I., Bengio, Y. 2016. Courville A., DeepLearning. Massachusetts Institute of TechnologyPress, Londra, 800s.
16. [16]Machine Learning Mastery. 2017. A GentleIntroduction to Transfer Learning for DeepLearning.https://machinelearningmastery.com/transfer-learning-for-deep-learning/ (Erişim Tarihi:06.05.2023).
17. [17]Kaggle. 2020. Dry Bean Dataset.https://www.kaggle.com/c/dry-beans-classification-iti-ai-pro-intake01. (Erişim Tarihi:22.06.2022).
18. [18]Koklu, M. 2020. Durum Wheat Dataset.https://www.kaggle.com/datasets/muratkokludataset/durum-wheat-dataset. (Erişim Tarihi:22.06.2022).
19. [19]UC Irvine Machine Learning Repository. 2020.Early Stage Diabetes Risk Prediction.https://archive-beta.ics.uci.edu/ml/datasets/early+stage+diabetes+risk+prediction+dataset. (Erişim Tarihi:22.06.2022).
20. [20]Rachakonda, L. 2020. Human Stress Detection.https://www.kaggle.com/datasets/laavanya/stress-level-detection. (Erişim Tarihi: 22.06.2022).
21. [21]UCI Machine Learning. 2020. MushroomClassification,https://www.kaggle.com/datasets/uciml/mushroom-classification. (Erişim Tarihi: 22.06.2022).
22. [22]Koklu, M. 2020. Pistachio Image Dataset.https://www.kaggle.com/datasets/muratkokludataset/pistachio-image-dataset. (Erişim Tarihi:22.06.2022).
23. [23]Koklu, M. 2020. Pumpkin Seeds Dataset.https://www.kaggle.com/datasets/muratkokludataset/pumpkin-seeds-dataset. (Erişim Tarihi:22.06.2022).
24. [24]Koklu, M. 2020. Raisin Dataset. https://www.kaggle.com/datasets/muratkokludataset/raisin-dataset. (Erişim Tarihi: 22.06.2022).
25. [25]Koklu, M. 2020. Rice Dataset Cammeo andOsmancik.https://www.kaggle.com/datasets/muratkokludataset/rice-dataset-commeo-and-osmancik.(Erişim Tarihi: 22.06.2022).
26. [26]Chollet, F. 2017. Deep Learning with Python.Manning Publications, New York, 384s.
27. [27]Kaggle. 2019. Stratified Shuffle Split workingwith less data.https://www.kaggle.com/viswanathanc/stratifiedshufflesplit-working-with-less-data. (ErişimTarihi: 22.06.2022).
28. [28]Medium. 2020. StratifiedKFold v.s KFold v.s.https://xzz201920.medium.com/stratifiedkfold-v-s-kfold-v-s-stratifiedshufflesplit-ffcae5bfdf.(Erişim Tarihi: 12.10.2020).
29. [29]Github. 2019. Cross-Validation: Why and how todo it. https://srikarvaka.github.io/model-evaluation/Cross-validation/. (Erişim Tarihi:22.06.2022).
30. [30]Koklu, M., Ozkan, I. A. 2020. Multiclassclassification of dry beans using computer visionand machine learning techniques. Computers andElectronics in Agriculture, 174, 105507.
31. [31]Kaya, E., Saritas, İ. 2019. Towards a real-timesorting system: Identification of vitreous durumwheat kernels using ANN based on theirmorphological, colour, wavelet and gaborletfeatures. Computers and Electronics inAgriculture. 166, 105016.
32. [32]Ergün, Ö. N., İlhan, H. O. 2021. Early stagediabetes prediction using machine learningmethods. European Journal of Science andTechnology Special, 29, 52-57.
33. [33]Rachakonda, L., Mohanty, S. P., Kougianos, E.,Sundaravadivel, P. 2019. Stress-Lysis: A DNN-Integrated edge device for stress level detectionin the IoMT. IEEE Transactions on ConsumerElectronics, 65(4), 474-483.
34. [34]Wibowo, A., Rahayu, Y., Riyanto, A., Hidayatulloh,T.2018. Classification algorithm for ediblemushroom identification. InternationalConference on Information and CommunicationsTechnology (ICOIACT), 06-07 March, Yogyakarta,250-253.
35. [35]Ozkan, I. A., Koklu, M., Saraçoğlu, R. 2021.Classification of Pistachio Species UsingImproved k-NN Classifier. Progress in Nutrition,23(2), e2021044.
36. [36]Koklu, M., Sarigil, S., Ozbek, O. 2021. The use ofmachine learning methods in classification ofpumpkin seeds (Cucurbita pepo L.). GeneticResources and Crop Evolution, 68(1), 2713–2726.
37. [37]Cinar, I., Koklu, M., Tasdemir, S. 2020.Classification of raisin grains using machinevision and artificial intelligence methods. GaziMühendislik Bilimleri Dergisi, 6(3), 200-209.
38. [38]Cinar, I., Koklu, M. 2019. Classification of ricevarieties using artificial intelligence methods.International Journal of Intelligent Systems andApplications in Engineering, 7(3), 188–194.